Personalized Deep Learning Robot to Help in Autism Therapy

Children suffering from autism spectrum condition, they often fail to recognize emotional states of people around them. They usually fail to distinguish between happy, sad, and fearful faces. As a solution, some therapists suggest kid-friendly robots to demonstrate emotions and encourage children in imitating these emotions and responding to them in suitable ways.

This type of therapy works best if the robot works smoothly to keenly observe the child behavior during the therapy. Researchers at the MIT Media Lab developed a personalized machine learning program that estimates the interest of a child and his interactions. Personalization of robot improves the response of child and assessments by human experts, with a correlation score of more than 60 percent.

These will be the challenge for human observers to reach high levels of child engagement and behavior. Their correlation scores were observed at between 50 to 55 percent. These robots are trained under human observation and during tests, they were consistent.
The researchers are focused to create robots to collect personalized information and encourage interactions between robot and children suffering from autism, said Oggi Rudovic one of the researchers at Media Lab. Challenge of creating artificial intelligence (AI) and machine learning to work on autism is a long process, as AI requires large data. In case of autism, if any data mismatches, then the normal AI approves as fail. The team has using AI in other areas such as pain monitoring and prediction of Alzheimer’s diseases.

This therapy proceeds as human therapists show a child photo, flash, and faces. Then therapist robot encourages a child to recognize the expressions. The child’s behavior provides valuable feedback and next lesson they have to take them to human therapists.